The Perils of Predicting Proliferation
Alexander H. Montgomery and Scott D. Sagan
Word count: 12,319
1
The systematic study of the causes and consequences of the spread of nuclear weapons is
at an early stage of development. The essays in this special issue (Gartzke and Kroenig) signal a
renaissance of interest in quantitative empirical studies of nuclear proliferation, an approach that
was mostly dormant after early attempts at predicting proliferation in the 1980s (Kegley 1980;
Meyer 1984). A new generation of political scientists, however, has recently taken up the charge
of assessing the causes and consequences of the spread of nuclear weapons (Singh and Way
2004; Asal and Beardsley 2007; Jo and Gartzke 2007; Fuhrmann 2008). The authors in this issue
continue that important effort by creating new datasets to widen the scope of inquiry, using new
and innovative methodologies, and testing a variety of theories to explain how and why states
acquire nuclear weapons and to predict the effects of the spread of nuclear weapons on the
likelihood and outcome of military conflicts.
This article reviews and critiques this second wave of quantitative literature on nuclear
weapons proliferation. These new studies provide an important counterpart to the mainstream
proliferation literature, which generally focused on in-depth individual case studies or small-n
cross-case variation rather than attempting to test hypotheses in a systematic way across all states
over long periods of time. The papers in this issue demonstrate the advances made since early
attempts at quantifying the causes and effects of proliferation, significantly improving our
understanding of these phenomena. This new research on the causes of proliferation usefully
focuses on the causes of nuclear assistance from other countries and estimates the effects of that
assistance on nuclear weapons acquisition. They demonstrate convincingly that general nuclear
technology assistance is affected by patterns of enmity and alliance and that receiving sensitive
2
nuclear assistance (help with uranium enrichment, plutonium reprocessing, or nuclear weapon
design) significantly increases the chances of acquiring nuclear weapons. (Fuhrmann, Kroenig).
The articles in this special issue also offer innovative insights into the nuclear
optimism/pessimism debate: does the proliferation of nuclear weapons make states more
cautious and reduce the likelihood of war, or does it lead to nuclear-armed states seeking
additional gains and risking catastrophe? The findings are mixed. States that have nuclear
weapons are more likely to initiate Militarized Interstate Disputes (MIDs), and this tendency
increases with experience (Horowitz, Gartzke and Jo). However, one article argues that the
increase in MID initiation is due to endogeneity (Gartzke and Jo), while another demonstrates
that as states become more experienced with nuclear weapons, they become less likely to
reciprocate military challenges or have their challenges reciprocated (Horowitz). The first two
findings bolster the pessimism argument, while the latter two support the nuclear optimists’
position.
These articles also offer insights into the effects of nuclear weapons on the outcomes of
crises and conflicts. Here the authors find that dyads where only one state has nuclear weapons
are more likely to experience all levels of force in MIDs, with the nuclear state winning more
often and achieving a settlement more quickly in ICB (International Crisis Behavior) crises.
Dyads where both have nuclear weapons are much less likely to experience the maximum level
of force (War) in a MID, but are no more or less likely to win or to experience a shorter conflict
in an ICB crisis (Asal and Beardsley, Rauchhaus). Thus, nuclear states seem to have incentives
to be belligerent against both nuclear and non-nuclear states (which helps the pessimists' case),
refraining from escalating to war only against other nuclear states (which helps the optimists'
case).
3
These new studies of nuclear proliferation, however, despite using the most up-to-date
datasets and techniques available, do not solve five serious problems that have also plagued
earlier quantitative studies of nuclear proliferation. First, it is inherently difficult to have accurate
coding of the dependent variables regarding whether states are exploring, pursuing or have
acquired nuclear weapons: some governments’ civilian nuclear power programs reflect internal
ambivalence about whether the state should pursue a bomb option, and the intense secrecy
surrounding weapons programs has meant that earlier published data sets left out many cases of
covert programs or the creation of “hedging” options, a problem likely to remain persistent for
the foreseeable future. While the papers in this issue use a common dataset for nuclear weapons
possession, increasing the comparability of results across this set of papers, their use of a
common data set decreases the robustness of their findings because, as we will demonstrate,
other reasonable codings on when states initiated or completed nuclear weapons programs
produce very different results. Second, the coding rules used for existing independent variables
in nuclear proliferation analyses are also problematic. Too often quantitative researchers measure
what is easily measured rather than find ways of accurately capturing the concepts that our
theories suggest should be important. Regarding the causes of proliferation, for example,
important factors that have been discussed in historical case studies of proliferation such as
leaders’ psychology, bureaucratic power and military autonomy, and the desire for prestige are
often excluded altogether or measured poorly in statistical studies. The problem also exists for
the quantitative literature on the consequences of proliferation, in which the relevant independent
variable is assumed to be a state’s first nuclear weapon, despite a long-standing debate in the
historical literature about whether states are deterred by virtual nuclear weapons, a single bomb,
a small arsenal, or a second-strike nuclear capability. Instead of developing new data sets to test
4
for the effects of such factors, the literature uses proxy variables based on pre-existing data sets,
making the studies particularly vulnerable to the “looking for the key under the lamppost”
charge, as we will demonstrate below. Third, methodologies and data sets need to be tightly
coupled to the empirical questions under investigation. Including a broader set of states in a
dataset without proper controls allows states that cannot have any influence on processes or
outcomes to skew results, as we will also demonstrate. Fourth, some quantitative proliferation
articles have findings that are trivial, providing us with insights that we already knew or at least
believed to be true. While it is always valuable to provide tests of commonly held views to see if
they are indeed accurate, many of these studies have failed to produce counter-intuitive insights,
which would add more significantly to our understanding of proliferation. Fifth, and finally,
statistical findings can ignore or gloss over individual data points that are crucially important for
policymaking and wider scholarly debates. We note in this article, for example, that the 1999
Kargil war between India and Pakistan–the only undisputed case of a war between nucleararmed states is simply ignored by these studies, as it does not appear as a separate dispute in the
MID database, and is not listed as a war in the COW or ICB databases. 1
It is valuable in this light to compare the study of nuclear proliferation to the study of the
"democratic peace" in international relations. In both of these research areas, the scholarship has
slowly expanded from qualitative theoretical and historical inquiry to include more gametheoretic approaches and systematic quantitative testing of hypotheses. The development path of
research in the democratic peace has progressed significantly in the past two decades in ways
that are instructive. This literature began with deductions based on political theory and limited
historical evidence that suggested that democratic states rarely, if ever, go to war against each
other (Kant 1917 [1795]; Doyle 1986). Vigorous debates about the democratic peace theory
5
emerged in political science starting in the 1980s, and a progressive research program of many
scholars using multiple methods eventually produced a significant, though by no means
complete, degree of consensus about how and why democracies differ from non-democracies in
their foreign policy and international security behavior (Lake 1992; Maoz and Russett 1993;
Fearon 1994; Mansfield and Snyder 1995; Schultz 1998; Elman 1999; Russett and Oneal 2001;
Reiter and Stam 2002; Gartzke 2007). Our understanding of the effect of democracy on the
likelihood and conduct of war has been significantly enhanced by both rich historical and
rigorous quantitative scholarship on the democratic peace. We hope that this special issue and
this critique can lead scholars to emulate the democratic peace literature, produce more multimethod research, and contribute further to improved understanding of the causes and
consequences of nuclear proliferation.
In the first section, we review the earlier historical case-study and quantitative research
on nuclear proliferation and compare its findings with the more recent quantitative literature on
proliferation. Here we note improvements as well as lingering problems with the coding,
methodology, and empirical findings. Having established the context for the remainder of the
essay, we proceed by discussing the accomplishments and remaining problems in the essays in
this special issue. Here we analyze all three topics in this issue: 1) the supply and regulation of
nuclear technologies; 2) the effects of the spread of nuclear weapons on the likelihood of
conflict; and 3) the effects of nuclear weapons on the consequences of crises and conflict. We
conclude with a discussion of the implications for future research and policy debates about how
to deal with nuclear proliferation.
6
Previous Studies of Proliferation
Different studies of the causes of nuclear proliferation divide incentives and actors in different
ways. In a widely-cited analysis of nuclear nonproliferation, Scott Sagan compares three models:
security, domestic politics, and norms. The first focuses on military security motivations; the
second emphasizes domestic political or bureaucratic coalitions that form to support or oppose
nuclear weapons; and the third focuses on norms and prestige considerations that encourage
acquisition of nuclear weapons (Sagan 1996/97, 1999). He finds strongest support for the
security model, though he argues that domestic interests and prestige concerns are “sufficient,
but not necessary” conditions for proliferation in a limited number of cases. Scholarly debate
continued about how weigh these different motivations for proliferation, with some realists
arguing for the singular importance of security motivations (Mearsheimer 1990; Frankel 1993)
and other scholars focusing on the constraints stemming from domestic economic interests or
normative opposition against nuclear weapons (Solingen 1994, 1998, 2007; Katzenstein 1996).
Suzette Grillot, Alexander Montgomery, and others have, in contrast, argued that role
conceptions and identity play a major role in producing interests in starting and ending nuclear
weapons programs (Chafetz, Abramson, and Grillot 1996; Grillot and Long 2000; Montgomery
2005). Additional studies include cognitive and psychological factors that may contribute to
decisions to acquire the bomb (Ogilvie-White 1996, pp. 51-53; Hymans 2006).
Much of this historical case-study work, however, examined only cases in which
governments had developed nuclear arsenals. In order to overcome the limits of research that
“selects on the dependent variable,” scholars have also focused on specific cases in which
governments chose to get rid of nuclear weapons or forgo the option before they had acquired an
arsenal. T.V. Paul, for example, discusses the effects of power and norms, and finds that states'
7
security positions are primarily responsible for its willingness to accept nuclear disarmament
(Paul 2000). Some valuable case study-focused accounts include all of these factors without
seeking to analytically separate them into individual accounts (Reiss 1988; Reiss and Litwak
1994; Reiss 1995; Levite 2002/2003; Campbell, Einhorn, and Reiss 2004).
The first wave of quantitative studies sought to rectify the weaknesses of limited case
studies. Charles Kegley's original quantitative study attempted to measure twenty different
national and systemic measures for proliferation in a scattershot fashion, finding significant
variation in three categories: threat perception, achievement of international status, and domestic
satisfaction. His study is unique among quantitative studies in that it attempts to measure directly
a state's social role-conception in the international system rather than using a proxy for overall
status (Kegley 1980). Stephen Meyer divided incentives into three categories: political powerprestige, military-security, and domestic politics, then measured these using fifteen indicators.
He found that fear of a well-armed rival, pariah status, and desires for regional preeminence were
the most important proliferation incentives, but that alliances with a nuclear power, membership
in the NPT, and fear of pre-emptive strikes helped to dissuade states (Meyer 1984, pp. 44-74).
In the second, more recent, wave of proliferation studies, Sonali Singh and Christopher
Way divide incentives between internal (regime type and liberalization-related incentives) and
external (security threats and guarantees) determinants, while Dong-Joon Jo and Erik Gartzke
divide their incentives along Sagan's three models (Singh and Way 2004, pp. 862-865; Jo and
Gartzke 2007, pp. 168-171). Singh and Way find strong support for both enduring rivalries and
militarized disputes on the external side, and limited support for economic openness and
liberalization as domestic political factors. Jo and Gartzke find considerable support for status-
8
driven motives and somewhat less support for domestic political and international security
motives.
These studies also include supply-side factors that may limit proliferation, a subject that
is given less attention in the qualitative literature. Meyer developed a set of fourteen factors that
contribute towards a domestic capability, divided into factors that influence mining and
processing of nuclear materials, construction and operation of nuclear facilities, and weapons
fabrication; a subset of his indicators were extended through 1992 by Jo and Gartzke. Singh and
Way also analyze technological determinants using more general indicators, which are coded
into three variables: GDP, a general industrial threshold, and specific industrial capabilities.
(Meyer 1984 Appendix B; Singh and Way 2004, pp. 867-869; Jo and Gartzke 2006, pp. 5-7). All
three models find consistently strong support for supply-side factors as major determinants of
proliferation.
The Meaning of Proliferation
An important insight from these studies and their descendants is that more gradations exist
regarding the status of a state than simply being nuclear or non-nuclear. The contribution of the
quantitative proliferation literature here is important, but because quantitative studies of nuclear
proliferation by Meyer, Singh and Way, and Jo and Gartzke have adopted different methods of
coding nuclear activity, they are difficult to compare. Meyer simply coded whether a state had
made a "decision" to seek nuclear weapons or not, and did not examine whether states succeeded
or not. Jo and Gartzke operationalized this variable as a measure of whether the "highest decision
maker in a given state authorized [or ended] a nuclear weapons program" for official programs or
the year in which "nuclear activities increase noticeably" for clandestine programs. They also
added acquisition data for the dates on which "each state was ready to quickly assemble nuclear
9
components into nuclear weapons." Singh and Way measure nuclear status along a continuum
with four "degrees of nuclearness:" no interest, exploration, pursuit, and assembly. The first
degree is simply no interest in nuclear weapons. The second degree is exploration, which is
"demonstrated by political authorization to explore the option or by linking research to defense
agencies that would oversee any potential weapons development." To qualify as pursuing nuclear
weapons, states must take steps such as "a political decision by cabinet-level officials, movement
toward weaponization, or development of single-use, dedicated technology." Finally, both Jo and
Gartzke and Singh and Way code states as having fully acquired weapons if they either test or
possess a "functional nuclear weapon." (Meyer 1984 Appendix A; Singh and Way 2004, pp.
866-867; Jo and Gartzke 2006, pp. 1-2) The codings from Meyer, Singh and Way, and Jo and
Gartzke are compared in Table 1.
--TABLE 1 ABOUT HERE-The key idea that nuclear proliferation should not be viewed as a single end-state is
valuable. Proliferation is a process by which countries move closer to or away from different
thresholds toward developing the bomb. Countries will not necessarily stay solidly in one state of
“nuclear latency” or another, as internal and external conditions that fuel or suppress
proliferation may change over time. Governments do not actually need to 'decide' to 'go nuclear'
until the moment that they test a nuclear device; even then, the decision to develop a deliverable
weapon or declare nuclear weapons status can be independent of a nuclear test. The case of India
is instructive; the 1974 test device was declared a "Peaceful Nuclear Explosive" and India did not
claim nuclear weapons state status, nor did they weaponize for a number of years. In 1998, the
New Delhi government made two decisions: the first was to test; the second was to declare their
status as a nuclear weapons state.
10
The ambiguity of nuclear proliferation—and consequently, the pitfalls of selecting one
coding system or another—becomes apparent when comparing different coding systems for
exploration, pursuit, and acquisition of nuclear weapons. Even the dates of the programs for the
initial five nuclear states (N5) are debatable; Meyer has China starting in 1957, while Jo and
Gartzke start in 1956 and Singh and Way have China jumping directly to pursuit without actually
exploring nuclear weapons first in 1955. After the N5, the data quickly becomes quite murky.
Dates for Israeli acquisition range from 1966 to 1972 (with Meyer not even coding Israel as
deciding to seek nuclear weapons until 1968); Libya, Switzerland, Australia, Algeria, and
Taiwan's second attempt at a nuclear weapons program do not appear in Jo and Gartzke, while
Iran's first attempt in the 1970s is excluded from Singh and Way. Germany and Japan's World
War II programs are not included in the latter dataset as well.
Complete agreement between Jo and Gartzke/Singh and Way should not be expected, as
Jo and Gartzke attempt to code decisions made by officials, whereas Singh and Way look at the
capabilities of states. Variation in coding and datasets can be good if the coding rules and
decisions are matched up with the appropriate theories. Still, if coding is based on capabilities, as
it is for Singh and Way, then Japan and Germany might very well be coded as "exploring" an
option at least for the time before they signed or ratified the NPT, if not today; the histories of
these two nuclear-capable states are at least ambiguous, and some historical studies and schemes
for coding nuclear status list both as “hedging” rather than "not interested" (Mackby and
Slocombe 2004; Campbell and Sunohara 2004; Levite 2002/2003).. Yet if coding is based on
decision-making (Jo and Gartzke), Argentina and Brazil should probably not be coded as
pursuing nuclear weapons. Both developed uranium enrichment programs, yet the Brazilian
Navy's enrichment program was originally directed at fuelling nuclear power plants for
11
submarines (Barletta 1997), while Argentina’s was intended for civilian nuclear power plants
(Carasales 1999; Hymans 2001).
As can be seen in Table 1, Meyer's dataset, coded to end in 1980, missed many programs
– including Pakistan, Brazil, Iraq, Iran, Argentina, Sweden, Yugoslavia, Taiwan, and Romania –
primarily due to lack of data at the time. While we presumably have somewhat more confidence
in some of these estimates now, new information could change these historical or current
proliferation data sets at any point. For example, how should we code Syria, in light of the
evidence that a covert nuclear facility was destroyed in 2007 (Albright and Brannan 2008)? Even
if we had reliable and complete information on states' nuclear programs, however, the ambiguity
about states' programs is not confined to our perceptions and estimates; it is inherent in the
phenomenon itself. States do not necessarily make decisions “to go nuclear”; often, the meanings
of the programs are constructed by multiple actors within each state as the program progresses.
The sociology of nuclear proliferation argues that these programs are better understood by
examining the broader political and historical discourses of which nuclear science is a part
(Flank 1993-1994; MacKenzie and Spinardi 1995; Abraham 1998); proliferation is a contingent
historical phenomenon, not the inevitable result of the workings of the pressures of the
international system.
India is often offered as an example of what appears externally as ambiguity, which
internally might be better characterized as ambivalence. Itty Abraham (2006) argues that India
developed a nuclear program as part of a grand modernization scheme with the ultimate end of
legitimating its role as an independent, autonomous, self-sufficient postcolonial nation. To this
end, India remained "ambivalent" about its nuclear program. It adopted the seemingly
contradictory stances of leading the global charge for a robust nonproliferation regime and
12
championing the advances of its nuclear program, to the exclusion of all mentions of the foreign
aid it received in its development "Ambivalence," writes Abraham, "is a permanent feature of the
nuclear condition… a recognition of the inability to wholly control nuclear events." (p.56) As
nuclear programs evolve, bureaucracies are established and independent political interests
become entrenched. Beyond organizational pressures, however, ambivalence is a result of the
awareness that the potential uses for and demands of nuclear science can change over time. The
role that the program plays as an expression of state identity, too, can change as the relationship
between the state and larger normative-ideational structures changes. Along the road to nuclear
weapons, there are many potential paths; the meaning of specific nuclear technology programs is
therefore contingent, not set. Consequently, it is not simply that scholars or the CIA must
develop estimates of ambiguous “intent” on the part of potential proliferators; the process of
proliferation itself may often be better understood as “ambivalence” inside the governments
involved.
Comparing Methods and Results
In addition to the difficulties of defining proliferation, there are also methodological problems
with the second wave of quantitative proliferation studies that create further questions about the
comparability and validity of their conclusions. Both Singh and Way and Jo and Gartzke have
multiple stages of proliferation, but they deal with them in very different ways, and adopt very
different options for analysis. Contained in their methods are implicit assumptions that make
little substantive sense. In Singh and Way's analysis, even states that have no nuclear program
are included in their models of nuclear acquisition. Yet states can't acquire weapons without first
pursuing them,2 and it seems unlikely that states won't pursue nuclear weapons without first
exploring the option first, even if only for a short period of time. Jo and Gartzke effectively
13
count each and every year a state has nuclear weapons as a separate decision to go nuclear,
thereby placing excessive weight on states that went nuclear very early. We explore the results of
changing these assumptions below. We use these two datasets to demonstrate our first three
critiques of the literature: the problems with existing codings of nuclear programs; the lack of
proper proxies for important variables; and the need for tighter correspondence between
methodologies and empirical questions.
Due to their finely-grained coding, Singh and Way look at four different potential
transitions. Singh and Way have possible transitions from no nuclear weapons program to
exploration (0 to 1) or pursuit (0 to 2), from exploration to pursuit (1 to 2), and from pursuit to
acquisition (2 to 3). States that cancel their nuclear programs revert back to 0, with the exception
of India, which moves from acquisition (3) in 1974 to exploration (1) in 1975. Due to leftcensoring, the USA never enters any analysis, the USSR only enters into the analysis for
acquiring nuclear weapons, the UK enters only for moving to pursuit, and Japan and Germany's
World War II programs never show up. Libya, Pakistan, and China move directly from no
nuclear weapons program to pursuit, skipping exploration. The coding that excludes these
programs is potentially problematic, since it removes 5 of the 26 possible cases due to leftcensoring and 3 more from the analysis of exploration moving to pursuit.
Singh and Way try two different tacks for analysis: first, a hazard analysis of their three
different levels, then a multinomial logit. A hazard model for each level (1,2,3) with the pool
restricted to states at the previous level of development (0,1,2) is the most empirically relevant
way to analyze this data, since it models each level as a prerequisite for the next. We drop all
states not already in the first stage (exploring a program) from an analysis of the second stage
(pursuit of nuclear weapons) while adding a year of exploration for Libya, Pakistan, and China
14
before they began pursuing; this drops the number of data points from 5578 to 250.3 We further
correct for time by setting the first year of exploration as the first year of being at risk for
pursuing nuclear weapons;4 Once these changes are made, the only variables that remain
significant are related to gross domestic product (GDPpc and GDPpc squared jointly), dispute
involvement, and the percentage of democracies. Dropping all states that were not already
pursuing from the analysis of nuclear acquisition takes the number of data points from 5784 to
210, making all variables except industrial threshold statistically insignificant; the data is so
sparse that the algorithm doesn't even converge correctly. 5 Moving to a more empirically
relevant methodology thus undercuts a number of the results. Finally, Singh and Way omit
entirely measures of status or any effects that the nonproliferation regime might have; while
there are good methodological reasons for not simply putting in measures of NPT membership or
systemic acceptance (see below), eliminating them entirely from the analysis is problematic.
Jo and Gartzke, by contrast, only have two stages: nuclear weapons programs and nuclear
weapons possession. Unlike Singh and Way, they disregard relatively poorly-documented
suspected programs such as those in Algeria that did not progress very far. While this choice is
consistent with their coding rules, they nevertheless do exclude a number of empirically-relevant
confirmed nuclear weapons programs such as Libya, Australia, Taiwan’s second attempt, and
Switzerland. Moreover, their regressions end much earlier (1992) than Singh and Way (2000).6
Their methodology also differs from Singh and Way. They perform a probit analysis on
the complete dataset to determine the effects of the independent variables on starting a nuclear
program. However, unlike Singh and Way, states do not exit the risk pool for starting a nuclear
program once they begin; they remain in the pool the entire time. Consequently, the results are
biased towards programs that exist for a longer period of time; The United States, for example, is
15
coded as having a nuclear program for 52 years, while North Korea only contributes 11 towards
the algorithm.
If Jo and Gartzke's analysis is corrected to eliminate this bias from the risk pool, several
very important substantive results change. In Table 2, we report the first differences that result.
We replicate their model for states starting nuclear programs in Model 1, then run the same
model while only counting the first year of nuclear acquisition rather than every year in Model 2,
then substitute the Singh and Way dates for exploration in Model 3.7 Economic capacity flips
sign in both Models 2 and 3, and becomes significant in Model 2; an increase in capacity by one
standard deviation decreases the chance of program onset by 71 percent. Diffusion becomes
insignificant in Model 2, then flips sign in Model 3. We suspect that this is due to improper
variable specification; instead of diffusion being based on direct connections to potential
suppliers (as some of the articles in this issue address), here it is simply based on time. Nuclear
threat flips sign and becomes insignificant, although it is now in the expected direction. The
effects of being a Major power increase drastically in Model 2, but become insignificant in
Model 3; being a Regional power no longer matters in Model 3 as well. However, we believe
that this is in part due to mismeasurement; rather than a measure of prestige, the measures of
status here are based on material power.
--TABLE 2 ABOUT HERE--
Jo and Gartzke usefully grapple with the notions of norms as potential constraints on
proliferation. They find that NPT membership (whether a state has ratified the treaty) “decreases
the likelihood of having nuclear weapons programs,” not a surprising finding. But Jo and
16
Gartzke also find that their NPT system variable (the proportion of NPT joiners to the total
number of states in the world) does not produce a statistically significant effect on nuclear
proliferation and conclude that this “indicates that the NPT has not curbed proliferation
incentives since the 1970s.”(p. 179). This last finding, however, is neither robust to alternative
measures, nor is it likely to represent a causal relationship. In Model 2 of our tests, the NPT
system variable is positive and significant, but becomes negative and insignificant in Model 3.
The NPT system effect measured in this manner, however, may just reflect the generally
increasing trend for more states joining the treaty regime through 1992.8 Moreover, the Gartzke
and Jo NPT membership coding fails to distinguish between states that join the NPT in order to
cement a bargain with regional rivals (where we would expect to see an effect), cheater states
that join the NPT with existing nuclear weapons programs such as Taiwan, Iraq, and Libya, and
states that join as a part of joining the world order and have no intent of ever developing a
civilian nuclear infrastructure, much less nuclear weapons, such as Bhutan, Papua New Guinea,
and Fiji. We should not then conclude that NPT status and norms are unimportant in curbing
nuclear proliferation, but rather that they have not been properly tested yet.
Finally, the sparsity of the data for nuclear acquisition is particularly troubling. For
example, although the official n is 250 and 210 for pursue and acquire respectively in Singh and
Way, in effect there are only 21 unique subjects with 15 failures for moving from explore to
pursue and only 15 subjects and 9 failures for moving from pursue to acquire. At these levels,
perhaps something like Qualitative Comparative Analysis, which uses fuzzy logic to explore
complex interactions between different variables in datasets with a low number of observations,
might be more appropriate (Ragin 2000) than traditional statistical analysis.
17
Progress in Prediction
The papers in this issue further this initial work by focusing on the supply and regulation of
nuclear technologies (Kroenig, Fuhrmann) and the effects of the spread of nuclear weapons on
the propensity of states to get into conflicts (Gartzke and Jo, Rauchhaus, Asal and Beardsley, and
Horowitz) as well as how nuclear weapons affect the dynamics and net outcomes of such
conflicts (Asal and Beardsley, Horowitz).
Methodologically, these papers have much to recommend them. They increasingly take
into account strategic interaction and endogeneity in various ways, through testing to see if
selection into crises affects their conduct and outcomes (Asal and Beardsley, Horowitz), and by
using instrumental variables for proliferation (Gartzke and Jo). They carefully include
sophisticated notions of the effects of time through splines (Gartzke and Jo, Horowitz), Taylorseries approximations (Asal and Beardsley), autoregression (Fuhrmann, Rauchhaus), or hazard
models (Kroenig, Asal and Beardsley). These are important methodological innovations. At the
same time, these papers replicate some of the errors of earlier work. We discuss how our first
four critiques apply to the papers below, then turn to the curious case of Kargil to elaborate on
the fifth.
Proliferation Determinants
Kroenig is the only author who directly extends upon the proliferation determinants literature,
adding a potentially significant cause of proliferation acquisition: sensitive nuclear assistance.
Kroenig’s coding of whether states received sensitive assistance or not (and from whom) is a
significant improvement over past efforts to determine the supply side factors influencing
proliferation. Kroenig fixes both of the major methodological problems present in previous
pieces, counting each nuclear weapons acquisition only once, and includes a censored model
18
with only states that are already pursuing a program. It is worth noting that in the censored
model, similar to the results above, very few of the independent variables are significant. In this
light, it is particularly important to note that sensitive nuclear assistance is still highly significant;
commendably, Kroenig tests his model against both earlier datasets as well as the consensus
dataset for this issue for robustness, although he does not report the full results.
What Kroenig does not highlight is also important: in robustness checks, he finds that one
particular type of civilian nuclear assistance (a state’s first reactor) and NPT membership
decrease the probability that a state will acquire nuclear weapons. The notion that the spread of
civilian technology will inevitably lead to more proliferation has received a lot of attention,
primarily due to a few important cases such as India, Iran, and North Korea. Furthermore,
Fuhrmann argues that defining civilian cooperation more broadly as nuclear cooperation
agreements (NCAs) does increase the probability of exploration, pursuit, and acquisition.
Kroenig offers an important counterpoint here in the form of evidence that both the NPT and
some civilian technologies do decrease the likelihood of weapons proliferation, providing some
evidence that the bargain of the NPT is holding, although our caveats regarding NPT
membership apply here as well.
Kroenig’s main findings are useful, but are hardly counter-intuitive. He concludes that
“states that have the ability to acquire nuclear weapons, either through international assistance or
domestic capability, are more likely to do so.” (p.3). It is valuable to know that important
sensitive assistance has been for new proliferants’ programs, but it is not surprising, to give the
most extreme example, that states that received bomb designs from other states were more likely
to develop a bomb than states that did not receive such designs.
19
While Kroenig focuses primarily on the effects of sensitive assistance, Fuhrmann adds to
our knowledge about who gives what kinds of nuclear assistance to whom, finding that states
make agreements with allies and with other states who share the same enemies. Fuhrmann,
uniquely among these papers, breaks down his analysis into several time periods, commendably
allowing for greater scrutiny of his results and giving a more accurate picture of the evolution
over time of assistance. Fuhrmann’s finding that membership in the NPT actually decreases the
probability that a state will receive civilian nuclear cooperation agreements (NCAs) is an
interesting and potentially troubling finding, given that this evidence seems to run counter to one
of the major incentives for states to join the NPT.
This result, however, while statistically sound, appears to be substantively misleading: it
is likely to be an artifact of the widespread growth and legitimacy of the NPT, since virtually all
states, including many that have no reason to sign such nuclear cooperation agreements, have
signed the treaty. Fuhrmann’s NPT result is strictly applicable to the period 1992-2000; this
outcome, then, likely comes from comparing the small number of states outside the treaty, who
generally are interested in nuclear technologies of all kinds but did not wish to submit to the
demands of the NPT, with the vast majority of states (84% in 1992) within the NPT, many of
whom have no interest in nuclear technology of any kind. In short, another interpretation of
Fuhrmann’s finding is that NPT member states that do not wish to acquire nuclear technologies
are less likely to get nuclear cooperation agreements than states who wanted nuclear technology
before they signed the NPT or the small number of states (India, Pakistan, and Israel) that have
remained outside the treaty. This is not a surprising finding. Fuhrmann also finds that “countries
exploring nuclear weapons are 107% more likely to receive nuclear aid for peaceful purposes”
than are countries not exploring a nuclear weapons option. (p. 27). This too appears to be a
20
startling finding (albeit one that decreases over time as well). Yet again it seems to be caused by
comparing states that are exploring nuclear weapons programs (all of which have civilian
programs as well), with states that have never even explored a nuclear weapons program, a large
number of which never had nuclear power programs at all.
Proliferation Consequences
The second major theme of this issue is the consequences of nuclear proliferation. The
previously existing scholarly debate on this subject is both theoretically rich and vigorous, but
deeply unresolved, in part because it revolves around making predictions about a highly
uncertain future. The nuclear optimists (Mearsheimer 1990; Waltz 1990; Mearsheimer 1993;
Karl 1996; Sagan and Waltz 2003 Ch.1, 3–4), following realist theory, envision a “perpetual
nuclear peace,” maintaining that states and statesmen behave with extreme rationality and will be
therefore be deterred from using nuclear weapons in crises or wars against other nuclear states.
They argue that overcoming preventive war incentives, developing second-strike capabilities,
and developing command and control systems is easy. Nuclear pessimists (Feaver 1992; Miller
1993; Sagan 1993, 1994; Feaver, Sagan, and Karl 1997; Sagan and Waltz 2003 Ch.2–3, 5; Sagan
2004) use organization theory and maintain that nuclear weapons are not controlled by states, but
rather by normal fragile humans in imperfect organizations. Their exploration of nuclear history
has uncovered much evidence suggesting that there were close calls to nuclear war during the
Cold War, and they predict therefore that new proliferators, especially those with poor command
and control systems, will be particularly dangerous nuclear weapons states.
There have been surprisingly few statistical studies, however, designed to determine how
common nuclear “near accidents” are or to learn how the acquisition of nuclear weapons
influences a states’ propensity to engage in conventional or sub-conventional war. The
21
exceptions focus on the success rate of extended deterrence and on the effects of nuclear
weapons in regional settings. Paul Huth, for example, finds that nuclear weapons increase the
likelihood of successful deterrence of attacks on one’s allies (Huth 1988, 1990). S. Paul Kapur
combines case-study research and statistical methods in his study of the effects of nuclear
weapons in South Asia and finds that nuclear weapons have led to an increase in the likelihood
of MIDs between India and Pakistan (Kapur 2007). But we lack broader studies of the
consequences of nuclear proliferation on different states’ war and crisis behavior.
The remaining papers in this issue help fill that gap and make significant contributions.
They turn the dependent variable in the causes of proliferation studies – a state’s nuclear
weapons acquisition – into the independent variable to assess the consequences of proliferation.
Unfortunately, although these papers use the same dates for when individual states acquired their
first nuclear weapon, increasing comparability, it is unclear what is the theoretically appropriate
threshold in a state’s nuclear weapons capabilities for nuclear weapons to have an effect on its
own or its rivals’ calculations and behavior. Even if the correct date of a state’s first nuclear
weapon could be definitively determined (which is not the case today for many new
proliferators), should we expect to see deterrent or compellent effects when a state is close to
getting a weapon, when it has one weapon, after it has acquired a small arsenal, or only once it
has a secure second-strike capability? An opaque proliferating state without a useable nuclear
weapon yet might still act more belligerently to encourage other governments to back down in
crises; similarly different adversaries to suspected proliferators may react quite differently to
challenges if they have different intelligence estimates of the opponent’s nuclear capabilities.
Take the secret Israeli nuclear program, for example: some scholars date Israel's nuclear
weapons capability from a covert 'decisive test' performed in November 1966, while others claim
22
a crude weapon was first assembled in or just on the eve of the 1967 war; yet the US thought that
Israel did not have nuclear weapons as late as 1969 (Cohen 2000, pp. 112-114; Montgomery and
Mount 2006, p. 18; Richelson 2006). When should one expect, therefore, that the USSR or Arab
states would be influenced by a possible Israeli nuclear weapon? Similar difficulties apply to
other states. India tested its 'Peaceful Nuclear Explosive' in 1974, which apparently was not
easily deliverable; according to some accounts; weaponization was not completed until 1988;
and this capability wasn't definitively demonstrated until the May 1998 nuclear tests. Dates for
Pakistan's capabilities range from 1987 to 1990, although again the May 1998 test date could
also apply. North Korea is even more enigmatic; the amount of plutonium that the North Koreans
after reprocessing a limited number of fuel rods by 1992 may not have been enough for even a
first weapon; further reprocessing in 2003 definitely gave North Korea a sufficient amount for a
few weapons, but this capability wasn't demonstrated definitively until October 2006. The
ambiguity of the status of the arsenal of South Africa in the 1970s and 1980s (Albright 1994, p.
43; Montgomery and Mount 2006, p. 10) and the inherited Soviet weapons in Belarus,
Kazakhstan, and Ukraine in the 1990s are further cases in point. Even using a first test date is
ambiguous; how should the suspected Israeli or South African test in 1979 or the October 2006
North Korean test (which many analysts believe was a failure) be coded?
A second common methodological problem in the consequences of proliferation literature
has been how to deal with selection effects and endogeneity. If nuclear states are found to win
more disputes than non-nuclear states is that because they are more powerful, or because they
cautiously enter only into disputes that they are likely to win? If states that acquire nuclear
weapons engage in more conflicts than non-nuclear states, are they really more conflict-prone
because they have nuclear weapons or do they have nuclear weapons because they are conflict
23
prone? The authors contributing to this special issue make important progress in dealing with the
selection effects and endogeneity problems. But they are less successful in dealing with the
difficulties of matching the coding independent variables to the theory being tested. Some of
their findings are also, as was the case with the causes of proliferation articles, quite obvious and
unlikely to overturn conventional wisdom.
Gartzke and Jo’s essay is a significant and creative contribution, assessing the effects of
the possession of nuclear weapons on militarized dispute (MID) onset and escalation. They find
two contradictory effects: while actual nuclear weapons possession is found to increase the
likelihood of initiating a dispute, when substituting an instrument (effectively, the probability of
nuclear weapons possession) instead to correct for possible endogeneity, they find no effect.9
Gartzke and Jo suggest that “nuclear weapons have no net effect on dispute propensity” because
some nuclear states become more aggressive and others more cautious and these effects wash
each other out. Moreover, they suggest that when facing a nuclear armed state, non-nuclear
target states acquiesce before a dispute reaches the dispute stage, but nuclear states escalate their
demands such that new disputes still occur. Again, the net effect of nuclear weapons acquisition
on dispute propensity will be minimal. This is an intriguing finding and their explanation is one
plausible potential cause of the results. We suggest two others. Possession of nuclear weapons
should deter disputes with non-nuclear states, but may not if a) a nuclear taboo exists and both
actors believe the dispute is below the threshold of plausible nuclear use (Paul 1995; Tannenwald
2007), or b) the non-nuclear initiator wishes to demonstrate that they will not be deterred by
nuclear weapons, as Stalin did at the beginning of the Cold War (Holloway 1996).
Gartzke and Jo have an additional intriguing finding – that “states with nuclear weapons
are more likely to receive diplomatic missions from other states than states without nuclear
24
weapons.” They interpret this result as evidence of and increased diplomatic influence that
nuclear weapons may bring to a state. This interpretation, however, is problematic for two
reasons. First, while having a higher number of connections to other states may result in
increased prestige or influence, whether this is the case depends heavily on whom you are
connected to and whether the connection is a positive one or seen as a necessary evil (HafnerBurton and Montgomery 2006; Hafner-Burton, Kahler, and Montgomery 2009). Second, Gartzke
and Jo conflate the legal status of official diplomatic recognition with the political influence –
arguing that “nations with demands need to be recognized before those demands can be
satisfied” – despite the fact that negotiations commonly exist between states that do not have
formal diplomatic recognition of each other. In short, diplomatic recognition, political influence
and international esteem are three very different things and it is difficult to avoid the impression
that this measure of “diplomatic influence” was chosen because it was readily available in the
COW data set. Gartzke and Jo usefully focus our attention on the connection between nuclear
weapons and diplomatic influence, but they have not, contrary to their claims, demonstrated that
“proliferators appear to prosper by becoming more influential diplomatically.”
Asal and Beardsley’s article “Winning with the Bomb” nicely follows on their early work
on the effect of nuclear weapons on crisis outcomes, using the ICB dataset (Asal and Beardsley
2007). The earlier work demonstrated that as the number of nuclear powers increases, the
violence level of a crisis decreases. Asal and Beardsley’s new article demonstrates that states that
have nuclear weapons have a bargaining advantage as well; the probability of victory (defined as
either outright victory or satisfaction with a compromise settlement) in a crisis increases by 14%
when a nuclear state faces a non-nuclear opponent in any crisis, and by 30% in “high salience”
crises. Moreover, Asal and Beardsley find that, controlling for other factors, the crises that
25
nuclear weapons states have with non-nuclear weapons states are significantly shorter on average
than are crises between two non-nuclear weapons states. They conclude that “nuclear weapons
provide more than prestige, they provide leverage. They are useful in coercive diplomacy and
this must be central to any explanation of why states acquire them.”
Upon further examination, however, the evidence for this conclusion is less impressive
than Asal and Beardsley acknowledge. Is a 14% increase in the probability of winning an
international crisis sufficiently high for statesmen to justify an investment that costs millions of
dollars and runs the risk of preventive war and potential opprobrium from the international
community? The higher “victory” percentage in high-salience asymmetrical nuclear weapon
cases is more impressive, but more than half of those victories are actually compromise
settlements.10 Moreover, the Asal and Beardsley salience measurement, while a commendable
attempt to measure the relevance of nuclear weapons to a crisis, is likely to be endogenous to the
conflict itself, since the presence of these weapons affects both the level of violence and the
stakes involved. Their salience definition, including both actual military violence and threats of
grave damage, strangely excludes cases in which no violence occurred or when only regional
threats existed even if nuclear weapons were highly relevant. Their coding thus leaves out many
of the Cold War and post-Cold War crises in which nuclear weapons use was contemplated or in
which threats to attack suspected nuclear weapons production facilities were made, including the
1954 Dien Bien Phu crisis, the 1961 Berlin crisis, the 1969 Usurri River crisis, the 1987 Brass
Tacks crisis, and the 1993-4 crisis over the North Korean nuclear program, and the 1999 Kargil
crisis (see below).
Finally, Asal and Beardsley include a robustness check to ensure that the results are not a
function of nuclear states selecting themselves into crises, which implies that nuclear states are
26
actually more likely to start crises in the first place. This offers empirical evidence that nuclear
states are more likely to enter into crises—even with other nuclear states. Yet it is not clear that
this finding is robust across different codings of nuclear status; indeed, an earlier iteration of the
same work with a much broader dataset of nuclear weapons states demonstrated that nuclear
states were less likely to enter into crises.11 A logical extension of this work would be to study
how the outcomes of asymmetrical crises influence the likelihood of further nuclear
proliferation; if nuclear states win asymmetric fights, how often do the losers then start nuclear
programs?
Rauchhaus’s essay -- “Evaluating the Nuclear Peace Hypothesis” – tests a long-standing
proposition in studies of nuclear deterrence theory and proliferation: the stability-instability
paradox. This paradox holds that while symmetrical possession of nuclear weapons may prevent
all-out war between two states, it may also increase the likelihood of low-level military conflicts,
because one or both of these states might believe that the other government would fear escalation
and accept defeat rather than move to higher levels of violence. The logic of the stabilityinstability paradox led analysts during the Cold War to worry that the USSR might be tempted to
use military force against US allies after Moscow had acquired nuclear weapons or a position of
nuclear parity. If both governments in a nuclear dyad believed that the other would back down
rather than escalate further in a conflict, however, this could lead to a dangerous game of chicken
with unpredictable consequences.
The value of Rauchhaus’ essay is not in its statistical finding that “when two states have
nuclear weapons, the negative coefficient indicates that they are less likely to go to war with one
another.” We knew that already and that result was perfectly predictable given his historical data.
Indeed, as he notes, “the dataset does not include any instances of two nuclear states engaging in
27
war with one another. Thus, there is a perfect correlation between the absence of war and dyads
where both states possess nuclear weapons.” The value of the essay is its sophisticated statistical
test of whether two nuclear armed states are more likely to use threats and military force at lower
levels of violence against each other once both have nuclear weapons. This essay provides
substantial proof of this proposition: MIDs, the use of force in MIDs, and fatal MIDs are all
more likely to occur in symmetrical dyads of nuclear states. Moreover, Rauchhaus carefully
controls for potential selection effects and provides a robustness check of his findings by using
an alternative coding of the nuclear weapons acquisition date variable. However, the
methodology used here (looking at undirected dyads) doesn't give as precise of a measurement of
crisis behavior than directed dyads would, since we cannot tell which state in an asymmetrical
dyad started the crisis.
Horowitz, unlike most of the literature on proliferation, takes into account the possibility
that states may change their behavior over time due to learning, and carefully assesses at who
initiates and reciprocates each dispute. His valuable article finds that "new" nuclear states are
slightly more likely to have their disputes reciprocated (and are much more likely to reciprocate
others' initial moves) than non-nuclear states, but that over time, this diminishes to the point
where experienced nuclear states are less than half as likely as non-nuclear states to have their
disputes reciprocated. “Mature” nuclear states are also slightly less likely to reciprocate others'
disputes than are non-nuclear states).
Horowitz robustly tests different codings for nuclear acquisition, finding that including
nuclear inheritors (Belarus, Kazakhstan, and the Ukraine) does not affect the results. Vicarious
learning, a particularly difficult-to-measure but potentially important concept, is partially
addressed in a robustness test. Presumably new nuclear states enter a world in which all states
28
have learned something about nuclear disputes, so that their learning curve might be less steep.
Future research could attempt to model vicarious learning as a diffusion process that occurs
proportional to the strength of network ties between states; e.g., the United Kingdom is likely to
learn lessons from the United States, but not from the Soviet Union; Iran might learn more
effectively from Pakistani experience with nuclear weapons than from the Israeli or US
experience.
These are valuable papers presenting intriguing findings. Further research will be needed,
however, to assess whether the causal mechanisms suggested in these studies are actually driving
the results. More statistical work will also be needed, as noted above, to provide more robust
confidence in the statistical findings. Future work, however, must be more careful than past
research in taking into account disconfirming evidence as noted below.
The Curious Case of Kargil
Statistical studies are a useful, indeed essential, method of testing social science theories about
how the world works and the articles in this issue have clearly helped generated new insights
about the sources of nuclear proliferation and its consequences. The use of quantitative research
methods, however, too often leads scholar to ignore important disconfirming evidence or to treat
it as “measurement error.” Such disconfirming evidence should instead lead to questioning about
the accuracy of the initial test of a theory or research about why there may be exceptions to the
rule. Consider, for example, the curious case of Kargil.
The Kargil war of 1999 is curious in two ways: first, because it offers apparently
damning evidence against both the democratic peace theory and the nuclear peace theory;
second, because quantitative scholars on both subjects have largely ignored the implications of
the Kargil war. The war occurred in the spring of 1999, when Indian armed forces discovered
29
that Pakistani northern infantry soldiers, disguised as Mujahideen guerilla fighters, had crossed
the Line of Control in Indian-held Kashmir and had taken up fortified positions in the mountains
above the town of Kargil. India and Pakistan were clearly nuclear weapons states after their 1998
nuclear tests. The 1999 conflict should be coded as an interstate war, since the most reliable
estimates of the fatalities in Kargil, from the Kargil Review Committee (2000), set the number of
Indian dead at 474 (p.23) and the “lowest estimate of regular Pakistani Army casualties is 700
killed,” (p.98) for a conservative minimum of 1,174. This is well above the 1,000 battle deaths
for war criteria used in the COW and other data sets. Kargil is an exception to the democratic
peace theory as well because India and Pakistan score a +9 and + 7 respectively in the Polity IV
data set for 1999 (Marshall and Jaggers 2003). Still, all the papers in this issue have ignored this
fact and its implications for the robustness of the tests about the consequences of proliferation. A
leading work in the democratic peace literature does at least address the Kargil war: Russett and
Oneal recognize Kargil as an apparent exception to the theory, but incorrectly suggest that it may
not count as a war because many of the deaths were “Islamic guerrillas, not regular Pakistani
troops” (Russett and Oneal 2001, p. 48).
It would be better to analyze apparent exceptions to the rule to develop alternative
theories and hypotheses for future research. In the case of Kargil, the Pakistani decision to send
troops into Indian-held Kashmir was apparently taken by military leaders, with minimal
involvement of the democratically-elected prime minister. (Bennett-Jones 2002, pp. 102-103;
Sagan 2009). This leads to a testable hypothesis that military-run governments or civilian
governments with inadequate operational control over the military may behave differently
regarding nuclear weapons and war initiation. The fact that the Pakistani military also opposed
the withdrawal from Kashmir and the cease-fire settlement in 1999 also suggests that such
30
governments may behave differently regarding nuclear weapons and escalation in crises and
conflicts (Sagan and Waltz 2003, p. 97; Musharraf 2006, p. 96). Future qualitative and
quantitative research will be needed to test these hypotheses.
Implications
This revival of the quantitative study of proliferation is a welcome advancement in nuclear
proliferation studies. Some of the questions and debates in this subfield can only be fully tested
when using statistical methods. Others will require mixed methods, combining historical case
studies, deductive reasoning, and quantitative research. These papers constitute significant
progress in our understanding of the causes of nuclear proliferation, the propensity of states to
initiate disputes, and the outcomes of those disputes. As with much good research, they provoke
as many questions as they answer. Here we have noted some of the puzzling and potentially
contradictory results, offer some tentative explanations, and propose an agenda for further
research.
The findings of the papers on the causes of proliferation usefully focus attention on the
technology assistance to potential proliferators. The next wave of research will need to
disaggregate the kinds of assistance much more than Fuhrmann and Kroenig do in their articles.
While it is not surprising that designs for nuclear weapons make proliferation more likely, it
would be helpful to know, for example, whether uranium enrichment or plutonium reprocessing
were more commonly abused by covert proliferators, including states that started weapons
programs and did not complete them. It would also be helpful to differentiate between NCAs
signed and those that were actually implemented, to avoid giving inappropriate weight to cases,
like the German NCAs with Brazil, which were not in the end honored and thus could not
contribute to nuclear weapons programs or acquisition (Squassoni and Fite 2005).
31
Disaggregation of different types of NCAs would also be useful; nuclear assistance with medical
devices and light-water power reactors are unlikely to aid nuclear weapons programs, while other
types of civilian assistance are much more contentious. The relationship between the NPT,
covert nuclear weapons programs, and civilian nuclear assistance also needs to be examined
more thoroughly before jumping to negative conclusions about the treaty. Further historical and
statistical study is required to unravel some of these puzzles: Why have different states have
joined the treaty and how has the treaty influenced their subsequent behavior (Sasikumar and
Way 2007)? How do different technologies affect proliferation? How much does sensitive
assistance decrease the time to acquisition of nuclear weapons? With the potential for a dramatic
increase in the use of nuclear power expected in the future, these questions have an immediate
policy-relevance.
The articles on the consequences of proliferation papers also point to the need for further
research. One important research agenda would focus more on initiation of conflicts to learn if
possession of nuclear weapons makes states more likely to initiate or be the target of disputes. In
their raw results, all the articles on proliferation consequences suggest that possessors are more
likely to be in crises, although only two have specific results for initiation (Gartzke and Jo and
Horowitz). Gartzke and Jo posit that increased initiation is not due to nuclear possession itself,
but Horowitz’s results also suggest possessors become more likely to initiate disputes with
experience. This is a very bleak conclusion for nuclear optimists: while nuclear weapons states
may have their challenges reciprocated less often over time, they may issue more challenges.
Future research could usefully focus on whether all states fit into that pattern, since some types
of regimes or leaders are more likely than others to be aggressive if they get nuclear weapons.
While Asal and Beardsley and Horowitz have taken important first steps in understanding the net
32
effect of nuclear weapons on conflict and bargaining outcomes, additional research is clearly also
required here. If initiation is actually not affected (or increased) by nuclear weapons, this should
have an effect on the bargains that these states get with or without crises or threats of force.
Future work in this important area of research could focus less on the overall trends and more on
the outliers: for example, are there certain states or certain kinds of regimes that do not become
more prudent over time after getting nuclear weapons? Finally, a useful follow-on study
regarding the stability/instability paradox would be to combine Rauchhaus’ statistical methods
and historical case studies to examine directly the dynamics of escalation from the threat to the
use of force and from uses of force to fatal MIDs to determine whether the fear of escalation by
one side or both determined the governments’ behavior or whether other factors were more
important. That research could provide insights into how confident we should be about the
likelihood that the “nuclear peace” will continue to exist in the future if more states acquire
nuclear weapons.
The exclusion of important theories of proliferation (or their inclusion with inadequate
measures) also tempers some of the conclusions reached here, and requires additional research.
Scholars have grappled for a long time with the notions of prestige, norms, social roles, and
psychology as motivations for the causes of proliferation. Yet since Kegley, none of the
quantitative articles on the causes of proliferation have successfully incorporated these theories
into their analysis. This affects both papers on the causes the likelihood of proliferation and its
the consequences, as the reasons for proliferation could potentially have significant effects on
how nuclear weapons are used once they are acquired. Difficult-to-measure concepts such as
civil-military relations and bureaucratic power and autonomy have also rarely been incorporated
into quantitative analyses,12 or when incorporated have been measured using existing measures
33
of regime type, which were developed for other purposes. Yet civilian control of the military and
central government control over nuclear scientists could have important effects on both the
likelihood of proliferation and its consequences .
Finally, the results that quantitative studies produce do not necessarily make for sound
policy advice; what may be true across an entire population does not necessarily make good
sense for dealing with individual cases. The finding that civilian nuclear assistance may often
lead to proliferation, for example, should not lead scholars to oppose the Clinton and Bush
administrations’ agreements to give light water-moderated nuclear power reactors to North
Korea, for that policy option is conditional on North Korea dismantling its more proliferationprone graphite-moderated reactor and returning its weapons-grade plutonium stockpile to the
IAEA. The observation that nuclear-armed dyads are less likely to go to war against each other
similarly should not necessarily lead to the prescription that more nuclear states would be better,
since the possible destruction of a full-scale war between nuclear powers could outweigh the
benefits of decreased likelihood of conventional conflict.
The curious case of Kargil underlines this final lesson for both the democratic peace and
the nuclear peace: rare does not mean never. This is less damaging for the democratic peace; as
long as democracy decreases war, there is still a net benefit that results from an increase in the
number of democratic states. But if nuclear weapons do not eliminate the possibility of nuclear
war entirely, their proliferation could lead to a disastrous outcome over time. While nuclear
optimists may therefore take some comfort in the results of some of these articles, nuclear
pessimists will remind us that unless the probability of nuclear war is zero, the potential
consequences of nuclear proliferation may still outweigh the benefits.
34
Table 1: Nuclear Status Coding Schemes
Meyer
(1942–1980)
Decide
Jo and Gartzke
(1941–2002)
Programs
Singh and Way
(1945–2000)
Explore
Possession
Pursue
Acquire
Acquisition
USA
1942–
1942–
1945–
*
*
1945–
Russia
1942–
1943–
1949–
*
1945–
1949–
United Kingdom
France
1947–
1956–
1941–
1954–
1952–
1960–
35
1945–
1946–
1947–
1954–
1952–
1960–
China
Israel
India(1)
India(2)
1957–
1968–
1964–1966
1972–
1956–
1955–
1964–1965
1972–
1964–
1966–
1988–
1971–1990
1979-1991
1955–
1949–
1954–
1975–
1955–
1958–
1964–
1980–
1964–
1972–
1974–1974
1988–
South Africa
1975–
1969–
1974–
1979–1993
Pakistan
Pursuit
Korea, South
1972–
1972–1975
1987–
1971–1975
36
1972–
1972–
1959–
1970–1978
1990–
Libya
Brazil
1970–
1953–
1978–1990
1970–
1978–1990
Korea, North
1982–
1965–
1980–
Iraq
1973–2002
1976–
1982–
1984–
1985–
Iran(1)
1974–1978
Iran(2)
1984–
Argentina
1976–1990
1968–
1978–1990
37
–
Exploration
Germany
1941–1945
Japan
1943–1945
Switzerland
Sweden
1946–1969
1954–1969
1946–1969
Yugoslavia(1)
1953–1963
1954–1965
Yugoslavia(2)
Australia
1982–1987
1974–1988
1956–1973
38
6–1973
Taiwan(1)
1967–1976
1967–1977
Taiwan(2)
1987–1988
Algeria
1983–
Romania
1981–1989
1985–1993
39
*Before initial observation in dataset.
40
Table 2: Percentage change in the corrected likelihood of a Nuclear Program (Jo and
Gartzke data, bold = p<.10)
Model 3:
Singh and
Way Onset
Category
Model 1:
Replication
Variable (increase)
41
Model 2:
Onset
Opportunity
Latent production capability (+1)
256.7%
258.0%
153.9%
International
Domestic
Politics
Security
Conventional
Domestic unrest
threat
(+1(+1
SD)SD)
-19.3%
219.1%
101.8%
301.7%
106.5%
179.9%
42
Democracy (+10)
-24.2%
Norms
-70.0%
154.1%
NPT membership
system (+35 (+1)
states)
-18.6%
-50.2%
-19.3%
-50.3%
208.8%
-17.7%
Status
Major power (+1)
2384.9%
43
11365.0%
1464.4%
Regional power (+1)
1332.3%
970.5%
248.8%
4697
4273
4059
N
Bibliography
Abraham, Itty. 1998. The Making of the Indian Atomic Bomb: Science, Secrecy and the
Postcolonial State. London, UK: Zed Books.
———. 2006. The Ambivalence of Nuclear Histories. Osiris 21 (1):49-65.
Albright, David. 1994. South Africa and the Affordable Bomb. Bulletin of the Atomic Scientists
50 (4):37-47.
Albright, David, and Paul Brannan. 2008. The Al Kibar Reactor: Extraordinary Camouflage,
Troubling Implications: Institute for Science and International Security.
Angrist, Joshua D., and Alan B. Krueger. 2001. Instrumental Variables and the Search for
Identification: From Supply and Demand to Natural Experiments. Journal of Economic
Perspectives 15 (4):69–85.
Asal, Victor, and Kyle Beardsley. 2007. Proliferation and International Crisis Behavior. Journal
of Peace Research 44 (2):139-155.
Barletta, Michael. 1997. The Military Nuclear Program in Brazil: Working Paper, Center for
International Security and Cooperation, Stanford University.
Beardsley, Kyle, and Victor Asal. 2007. Winning with the Bomb. Paper read at 103rd Annual
Meeting of the American Political Science Association, August 30-September 2, at
Chicago, IL.
Bennett-Jones, Owen. 2002. Pakistan: Eye of the Storm New Haven, CT: Yale University Press.
Campbell, Kurt M., Robert J. Einhorn, and Mitchell Reiss, eds. 2004. The Nuclear Tipping
Point: Why States Reconsider Their Nuclear Choices. Washington, DC: Brookings
Institution Press.
Campbell, Kurt M., and Tsuyoshi Sunohara. 2004. Japan: Thinking the Unthinkable. In The
Nuclear Tipping Point: Why States Reconsider Their Nuclear Choices, chapter 9, pages
218-253, edited by K. M. Campbell, R. J. Einhorn and M. Reiss. Washington, DC:
Brookings Institution Press.
Carasales, Julio C. 1999. The So-Called Proliferator that Wasn't: The Story of Argentina's
Nuclear Policy. Nonproliferation Review 6 (4):51-64.
Chafetz, Glenn, Hillel Abramson, and Suzette Grillot. 1996. Role Theory and Foreign Policy:
Belarussian and Ukranian Compliance with the Nuclear Nonproliferation Regime.
Political Psychology 17 (4):727-757.
44
Cohen, Avner. 2000. Nuclear Arms in Crisis under Secrecy: Israel and the Lessons of the 1967
and 1973 Wars. In Planning the Unthinkable, pages 104-124, edited by P. R. Lavoy, S. D.
Sagan and J. J. Wirtz: Cornell University Press.
Doyle, Michael W. 1986. Liberalism and World Politics. American Political Science Review 80
(4):1151-1169.
Elman, Miriam Fendius. 1999. Paths to Peace: Is Democracy the Answer? In The Use of Force:
Military Power and International Politics, 5th ed, pages 608, edited by R. J. Art and K. N.
Waltz. Lanham, MD: Rowman and Littlefield.
Fearon, James D. 1994. Domestic Political Audiences and the Escalation of International
Disputes. American Political Science Review 88 (3):577-592.
Feaver, Peter D. 1992. Command and Control in Emerging Nuclear Nations. International
Security 17 (3):160-187.
Feaver, Peter D., Scott D. Sagan, and David J. Karl. 1997. Proliferation Pessimism and Emerging
Nuclear Powers. International Security 22 (2):185-207.
Flank, Steven. 1993-1994. Exploding the Black Box: The Historical Sociology of Nuclear
Proliferation. Security Studies 3 (2):259-294.
Frankel, Benjamin. 1993. The Brooding Shadow: Systemic Incentives and Nuclear Weapons
Proliferation. In The Proliferation Puzzle: Why Nuclear Weapons Spread and What
Results, pages 37-78, edited by Z. S. Davis and B. Frankel. Portland, OR: Frank Cass.
Fuhrmann, Matthew. 2008. Exporting Mass Destruction? The Determinants of Dual-Use Trade.
Journal of Peace Research 45 (5):633-652.
Gartzke, Erik. 2007. The Capitalist Peace. American Journal of Political Science 51 (1):166-191.
Grillot, Suzette R., and William J. Long. 2000. Ideas, Beliefs, and Nuclear Policies: The Cases of
South Africa and Ukraine. Nonproliferation Review 7 (1):24-40.
Hafner-Burton, Emilie M., Miles Kahler, and Alexander H. Montgomery. 2009. Network
Analysis for International Relations. International Organization.
Hafner-Burton, Emilie M., and Alexander H. Montgomery. 2006. Power Positions: International
Organizations, Social Networks, and Conflict. Journal of Conflict Resolution 50 (1):3-27.
Holloway, David. 1996. Stalin and the Bomb: Yale University Press.
Huth, Paul K. 1988. Extended Deterrence and the Outbreak of War. American Political Science
Review 82 (2):423-443.
———. 1990. The Extended Deterrent Value of Nuclear Weapons. Journal of Conflict
Resolution 34 (2):270-290.
Hymans, Jacques E. C. 2001. Of Gauchos and Gringos: Why Argentina Never Wanted the
Bomb, and Why the United States Thought It Did. Security Studies 10 (3):153-185.
———. 2006. The Psychology of Nuclear Proliferation: Identity, Emotions, and Foreign Policy.
Cambridge, UK: Cambridge University Press.
Jo, Dong-Joon, and Erik Gartzke. 2006. Codebook and Data Notes for ``Determinants of Nuclear
Weapons Proliferation: A Quantitative Model''.
———. 2007. Determinants of Nuclear Weapons Proliferation: A Quantitative Model. Journal of
Conflict Resolution 51 (1):167-194.
Kant, Immanuel. 1917 [1795]. Perpetual peace: a philosophical essay. Edited by M. C. Smith.
London, UK: G. Allen and Unwin.
Kapur, S. Paul. 2007. Dangerous Deterrent: Nuclear Weapons Proliferation and Conflict in South
Asia. Stanford, CA: Stanford University Press.
45
Kargil Review Committee. 2000. From Surprise to Reckoning: The Kargil Review Committee
Report. New Delhi: Sage Publications.
Karl, David J. 1996. Proliferation Pessimism and Emerging Nuclear Powers. International
Security 21 (3):87-119.
Katzenstein, Peter J. 1996. Cultural Norms and National Security: Police and Military in Postwar
Japan, Cornell studies in political economy. Ithaca, NY: Cornell University Press.
Kegley, Charles W. 1980. International and Domestic Correlates of Nuclear Proliferation: A
Comparative Analysis. Korea and World Affairs 4:5-37.
King, Gary, Michael Tomz, and Jason Wittenberg. 2000. Making the Most of Statistical
Analyses: Improving Interpretation and Presentation. American Journal of Political
Science 44 (2):347-361.
Lake, David A. 1992. Powerful Pacifists: Democratic States and War. American Political
Science Review 86 (1):24-37.
Levite, Ariel E. 2002/2003. Never Say Never Again: Nuclear Reversal Revisited. International
Security 27 (3):59-88.
Mackby, Jenifer, and Walter Slocombe. 2004. Germany: The Model Case, a Historical
Imperative. In The Nuclear Tipping Point: Why States Reconsider Their Nuclear
Choices, chapter 8, pages 175-217, edited by K. M. Campbell, R. J. Einhorn and M.
Reiss. Washington, DC: Brookings Institution Press.
MacKenzie, Donald, and Graham Spinardi. 1995. Tacit Knowledge, Weapons Design, and the
Uninvention of Nuclear Weapons. American Journal of Sociology 101 (1):44-99.
Mansfield, Edward D., and Jack Snyder. 1995. Democratization and the Danger of War.
International Security 20 (1):5-38.
Maoz, Zeev, and Bruce Russett. 1993. Normative and Structural Causes of Democratic Peace,
1946-1986. American Political Science Review 87 (3):624-638.
Marshall, Monty G., and Keith Jaggers. 2003. Polity IV Project: Political Regime Characteristics
and Transitions, 1800-2002: Integrated Network for Societal Conflict Research.
Mearsheimer, John J. 1990. Back to the Future: Instability in Europe after the Cold War.
International Security 15 (1):5-56.
———. 1993. The Case for a Ukrainian Nuclear Deterrent. Foreign affairs 72 (3):50-66.
Meyer, Stephen M. 1984. The Dynamics of Nuclear Proliferation. Chicago, IL: University of
Chicago Press.
Miller, Steven E. 1993. The Case against a Ukrainian Nuclear Deterrent. Foreign Affairs 72
(3):67-80.
Montgomery, Alexander H. 2005. Ringing in Proliferation: How to Dismantle an Atomic Bomb
Network. International Security 30 (2):153-187.
Montgomery, Alexander H., and Adam J. Mount. 2006. Misunderestimation: Explaining US
Failures to Predict Nuclear Weapons Programs. Paper read at 102nd annual meeting of
the American Political Science Association, August 31-September 3, at Philadelphia, PA.
Musharraf, Pervez. 2006. In the Line of Fire: A Memoir New York, NY: Free Press.
Ogilvie-White, Tanya. 1996. Is There a Theory of Nuclear Proliferation? An Analysis of the
Contemporary Debate. Nonproliferation Review 4 (1):43-60.
Paul, T. V. 1995. Nuclear Taboo and War Initiation in Regional Conflicts. Journal of Conflict
Resolution 39 (4):696-717.
———. 2000. Power Versus Prudence: Why Nations Forgo Nuclear Weapons, Foreign policy,
security and strategic studies. Montreal, Quebec: McGill-Queen's University Press.
46
Ragin, Charles C. 2000. Fuzzy-Set Social Science: University Of Chicago Press.
Reiss, Mitchell. 1988. Without the Bomb: The Politics of Nuclear Nonproliferation. New York,
NY: Columbia University Press.
———. 1995. Bridled Ambition: Why Countries Constrain Their Nuclear Capabilities,
Woodrow Wilson Center special studies. Washington, DC: Woodrow Wilson Center
Press.
Reiss, Mitchell, and Robert S. Litwak, eds. 1994. Nuclear Proliferation after the Cold War,
Woodrow Wilson Center special studies. Washington, DC: Woodrow Wilson Center
Press.
Reiter, Dan, and Allan C. Stam. 2002. Democracies at War. Princeton, NJ: Princeton University
Press.
Richelson, Jeffrey. 2006. Spying on the Bomb: American Nuclear Intelligence, from Nazi
Germany to Iran and North Korea. 1st ed. New York, NY: Norton.
Russett, Bruce M., and John R. Oneal. 2001. Triangulating Peace: Democracy, Interdependence,
and International Organizations. New York, NY: Norton.
Sagan, Scott D. 1993. The Limits of Safety: Organizations, Accidents, and Nuclear Weapons.
Princeton, NJ: Princeton University Press.
———. 1994. The Perils of Proliferation: Organization Theory, Deterrence Theory, and the
Spread of Nuclear Weapons. International Security 18 (4):66-107.
———. 1996/97. Why Do States Build Nuclear Weapons? Three Models in Search of a Bomb.
International Security 21 (3):54-86.
———. 1999. Rethinking the Causes of Nuclear Proliferation: Three Models in Search of a
Bomb. In The Coming Crisis: Nuclear Proliferation, U.S. Interests, and World Order,
edited by V. A. Utgoff. Cambridge, MA: MIT Press.
———. 2004. The Problem of Redundancy Problem: Why More Nuclear Security Forces May
Produce Less Nuclear Security. Risk Analysis 24 (4):935-946.
———, ed. 2009. Inside Nuclear South Asia. Stanford, CA: Stanford University Press.
Sagan, Scott D., and Kenneth N. Waltz. 2003. The Spread of Nuclear Weapons: A Debate
Renewed. 2nd ed. New York, NY: W.W. Norton.
Sasikumar, Karthika, and Christopher R. Way. 2007. Paper Tiger or Barrier to Proliferation?
What Accessions Reveal about NPT Effectiveness. Paper read at 103rd Annual Meeting
of the American Political Science Association, August 30-September 2, at Chicago, IL.
Schultz, Kenneth A. 1998. Domestic Opposition and Signaling in International Crises. American
Political Science Review 92 (4):829-844.
Sechser, Todd S. 2004. Are Soldiers Less War-Prone than Statesmen? Journal of Conflict
Resolution 48 (5):746-774.
Singh, Sonali, and Christopher R. Way. 2004. The Correlates of Nuclear Proliferation: A
Quantitative Test. Journal of Conflict Resolution 48 (6):859-885.
Solingen, Etel. 1994. The Political Economy of Nuclear Restraint. International Security 19
(2):126-169.
———. 1998. Regional Orders at Century's Dawn: Global and Domestic Influences on Grand
Strategy, Princeton studies in international history and politics. Princeton, NJ: Princeton
University Press.
———. 2007. Nuclear Logics: Contrasting Paths in East Asia and the Middle East Princeton
studies in international history and politics. Princeton, NJ: Princeton University Press.
47
Squassoni, Sharon A., and David Fite. 2005. Brazil as Litmus Test: Resende and Restrictions on
Uranium Enrichment. Arms Control Today, October.
Tannenwald, Nina. 2007. The Nuclear Taboo: The United States and the Non-Use of Nuclear
Weapons since 1945. Cambridge, UK: Cambridge University Press.
Tomz, Michael, Jason Wittenberg, and Gary King. 2003. Clarify: Software for Interpreting and
Presenting Statistical Results (2.1).
Waltz, Kenneth N. 1990. Nuclear Myths and Political Realities. American Political Science
Review 84 (3):731-745.
48
Notes
1
MID 3.10 does include Kargil as a dispute that had a maximum hostility level of war, but has it
as the end point of a dispute starting in 1993; see Kapur 2007, p. 19. COW 3.0 only has wars
through 1997, while ICB 8.0 lists Kargil as "serious clashes," not "full-out war."
2
The cases of Belarus, Kazakhstan, and Ukraine are excluded from both datasets, since the
temporary location of nuclear weapons on their soil resulted from the breakup of the Soviet
Union rather than from a domestic program.
3
Singh and Way footnote this variant in their analysis (p.875, fn.32); we believe that this is the
more empirically relevant question.
4
This corrects for entry and exit times of the risk pool. In Singh and Way's analysis, a state is at
risk of moving to the third stage (pursuing) nuclear weapons from its first date of entry into the
system. In this corrected analysis, states enter the risk pool for pursuit of nuclear weapons only in
their first year of exploring nuclear weapons.
5
We omit a full analysis here due to space constraints; STATA datasets and do-files for
replication are available on the JCR website.
6
Like the Singh and Way left-censoring, the right-censoring of this data at 1992 is potentially
important, but beyond the scope of this analysis.
7
We replicate their Table 1, Model 1 (Jo and Gartzke 2007, p. 178).We used Clarify 2.1 (King,
Tomz and Wittenberg 2000, Tomz, Wittenberg and King 2003) to analyze these differences; all
continuous variables were set to their means, while all categorical, dichomomous, and count
variables were set to zero (generally the median) for the baseline.
8
The NPT system variable is correlated 0.9131 with the year; the fraction of states in the system
increases rapidly from .47 in 1970 to .84 in 1992
49
9
While we critique Jo and Gartzke 2007 for including multiple observations for nuclear
possession, in the Gartzke and Jo paper, this model makes more sense in this context, since it is
attempting to produce an instrument that correlates with the presence of nuclear weapons, not
just with the initial acquisition. Nonetheless, we find the instrument to be potentially problematic
since probit and logit functional forms can lead to misspecification (see Angrist and Krueger
2001, p. 80), and important variables such as status are excluded from the instrument.
10
Similarly, in high-salience crises with symmetrical nuclear dyads, they record a 21% increase
in victory compared to a dyad with two non-nuclear weapon states; however, this is entirely due
to an increase in compromise settlements, as absolute victory actually decreases.
11
The APSA 2007 (Beardsley and Asal 2007) version of the paper had a negative coefficient on
Nuclear A and a small positive coefficient on Nuclear B. This dataset had earlier dates for India
(1974), Pakistan (1987), and North Korea (1992), and included Ukraine, Kazakhstan, and
Belarus from 1991 until 1996, 1995, and 1996 respectively.
12
One notable exception is Sechser 2004.
50